Scholarlink

<p>Spatial data mining aims to discover implicitly useful knowledge on spatial datasets. Co-location patterns play an essential role in spatial data mining as the goal is to find a set of features whose instances are nearby in spatial space prevalently. However, the exponential explosion of co-location patterns often disturbs users from analyzing results. The problem of filtering meaningful results from a large scale of co-location patterns with complex relationships remains to be addressed. To solve the issue, we propose an interactive co-location pattern post-mining framework in this paper, and an algorithm is designed based on this framework. In detail, we innovate the multi-granularity fuzzy clustering embedding for co-location post-mining, which reveals complex relationships between co-location patterns and improves the mining quality due to capturing the rich semantics inherent in patterns. We further utilize Markov’s inequality to reduce embedding dimensionalities, owing to the high-dimensional embedding vectors that may lead to the curse of dimensionality. Moreover, a new co-location pattern sampling method is designed to provide higher-quality samples for interaction. The algorithm is evaluated on both transactional data and spatial data, for wider applications. The experimental results on three real datasets show that our work is 2.03% higher on the F-score and 2.99% higher on the accuracy indicator at least, compared with the state-of-the-art algorithms. The code is available at: <a href="https://github.com/CrazySpy/FITTER">https://github.com/CrazySpy/FITTER</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Fitter: post-mining user-preferred co-location patterns interactively

  • 期刊论文

  • 发表日期：

    2025年11月14日

查看全文
  • Xiwen Jiang,
  • Lizhen Wang,
  • Peizhong Yang,
  • Hongmei Chen

摘要

Spatial data mining aims to discover implicitly useful knowledge on spatial datasets. Co-location patterns play an essential role in spatial data mining as the goal is to find a set of features whose instances are nearby in spatial space prevalently. However, the exponential explosion of co-location patterns often disturbs users from analyzing results. The problem of filtering meaningful results from a large scale of co-location patterns with complex relationships remains to be addressed. To solve the issue, we propose an interactive co-location pattern post-mining framework in this paper, and an algorithm is designed based on this framework. In detail, we innovate the multi-granularity fuzzy clustering embedding for co-location post-mining, which reveals complex relationships between co-location patterns and improves the mining quality due to capturing the rich semantics inherent in patterns. We further utilize Markov’s inequality to reduce embedding dimensionalities, owing to the high-dimensional embedding vectors that may lead to the curse of dimensionality. Moreover, a new co-location pattern sampling method is designed to provide higher-quality samples for interaction. The algorithm is evaluated on both transactional data and spatial data, for wider applications. The experimental results on three real datasets show that our work is 2.03% higher on the F-score and 2.99% higher on the accuracy indicator at least, compared with the state-of-the-art algorithms. The code is available at: https://github.com/CrazySpy/FITTER.